Antonette C.

The Pentagon's APFIT program announced more than $500 million in new awards this week, and for the first time the list includes software-only procurements, over $70 million worth. APFIT sits under the Research and Engineering directorate and exists to move tech from small, non-traditional, and venture-backed vendors into the services and combatant commands fast. It has delivered more than 100 capabilities since launch, and this round pushes total awards past $2 billion. Half of the inaugural software list is explicitly AI: the ARCHER decision-advantage platform for Pacific Command, Cybergenome AI binary analysis for Cyber Command, and SCEPTER AI-based course-of-action generation for Southern Command, alongside a threat planning tool for the Air Force, a rapid intelligence-disclosure system for Spacecom, and a tactical assault kit for SOCOM. Each award runs $10-17 million. Those are modest sums next to the hardware in the same round

For everyone who cares about the quality of AI agents and the eval scores used to measure them, we're sharing some surprising results on why production evals need to be treated like the adversarial cat-and-mouse scenario normally reserved for hackers. The results will be made public in Las Vegas during our BSides session on "Breaking BOTS II: How Frontier AI Cheats Evals”. This builds on our research behind “Breaking BOTS I”, where we made the first successful agentic speed run to break the annual Splunk CTF. To our surprise, commercial AI models were not directly beating the evals, but something much more sneaky: they were getting the right answers by hacking the benchmark system. Our production AI team mostly encounters AI safety concerns when preventing human attackers from performing prompt injections, so this new kind of attack was a shock. Internal quality control systems were being subverted without a single malicious prompt. Worse, the attacks are coming from the AI itself, without any human attacker initiating them, and keep getting more tricky as frontier models and harnesses keep improving. Protecting against these attacks shifts how we build, and it’s hard to shrug them off when evals are so core to our trust in agentic AI systems for security, fraud, IT operations, and other important areas. DM us to connect if you're going to be in Vegas for Black Hat / BSides / DefCon / Ai4 – it’ll be a busy week for AI! Also just announced, we’ll be running two free 4-hour BSides trainings on "From Copilot to Commander: Building Agentic AI for Security Investigations", and “Tour de Graph: Logs to graph training”, in addition to the summer cohort of our 2-day masterclass on the same. Dates and details in the first comment Dave Herrald Roberto Rodriguez Amin Karbasi David Bianco Dawn Song  Ramakrishna Rao (Coach Ram) Ram Shankar Siva Kumar 🦝 Sarah Bird #BSidesLV  #defcon #blackhat  #ai4

A year ago from last month I lost the closest person to me in my entire life. In the heartache I learned some lessons: - never take those who love you unconditionally for granted - cherish every minute you get with those who truly see you - if someone else cares so deeply for you, trust you should care about yourself too - if you lose someone who truly cared for you, honor their dying wishes - never assume just because someone is young that there is time You will make friends in life. You will meet people who become a greater part of your every day. If you have someone who truly accepts everything about your being - your investment in them and their investment in you matters more than your career. My cousin Jeromy was like a brother to me. Since 3rd grade when we met at a water park where he almost drowned. The life guard didn’t get to him, my sister and I did and after pulling him out of the wave pool we spoke every day on the phone for most of my life a minimum of 2 hours. We always joked that the wave pool incident must have been fate. “We were born to protect each other through this life.” In a difficult way, Jeromy’s final words and dying wishes protected a future he would want for me. As crazy as it sounds it would sometimes be 8 hours on the phone that we’d talk. Every major moment of my life he was there, until his later days. I know for me there was a lot I took for granted. When someone accepts you regardless of your own flaws, it sends a message. It’s okay to be who you are. It doesn’t matter if others don’t get you. Somewhere along the way I forgot myself. Over the past year I have had to face myself in ways that business or a career can’t reckon with. I cannot be the person I used to be, I’ve forever changed though not in the ways I’ve preferred. I spent my life building toward a retirement for more than just me. Along the way I lost myself. I urge you to look at those who care for you in your own life. They matter more than your career ever will. The few who love you unconditionally will show you what truly matters in life. My businesses thrive today. Yet my heart grows somber. If anything I will honor Jeromy’s dying wishes, he hoped that I don’t give up on my potential. You shouldn’t dismiss your own potential. Perhaps his lesson is something I can share here for all who read. The last time I saw him he had lost his own father. The last time I spoke to him he said I finally relate to his pain… He lived an isolated life in his sickness in his final days. He harbored no hard feelings toward me. The hardest lessons in life come with the reckoning of losing a loved one. Not everyone will understand these feelings today. Inevitably we will all face the reckoning of losing a loved one. That day will show you what really matters in life. #dare2try

I'm speaking at the Black Hat Innovators and Investors Summit. Well, sort of. I'm actually emceeing the event, and following in the shoes of my friend Mike Privette who's done it previously. It's been awesome to watch the agenda put together by the Innovators and Investors Summit Advisory Board in Coleen C., Robert Hansen, Jon Oberheide, Ty Sbano, Michael Sutton and Prima Virani I'll also be emceeing the Black Hat Startup Spotlight competition, something amazing leaders such as Caleb Sima have done in the past. Safe to say, I consider this a real honor, and I hope to do the role justice and give back to the cybersecurity community that has given me so much and that I'm so passionate about. I love watching the intersection of founders, innovators, and investors, especially in the age of AI and its convergence with cybersecurity. This years Startup Spotlight Finalists have been announced as well: Perpetual Systems, Opnova, Mallory and Deception Check. The Startup Spotlight Judges include: ☁️ Trey Ford, Hollie Hennessy, Matthew Martin, Brooke M., Lucas Nelson, Sid Trivedi and Mikala Vidal Both events are shaping up to be incredible and I'm looking forward to seeing everyone there!

Exciting update for security folks heading to Las Vegas this summer: BSidesLV 2026 Training Grounds is hosting our highly-rated masterclass “From Copilot to Commander: Building Agentic AI for Security Investigations”. It builds on the most popular training from Black Hat 2025 that top fintech and federal hunting teams called the best AI 101 they’ve ever seen. Our founder Leo Meyerovich and former police Chief Inspector Sindre Breda will be teaching a *FREE* hands-on session on designing effective and reliable AI agents for vibes investigating, threat hunting, threat intel, investigation automation, and operations. Look forward to discussions and labs on LLM foundations, modern agent harnesses, and core SOC/IR workflows like contextualization, triage, and investigation. We’ll cover how to do this with your favorite LLM models, agents, databases/tools (SPL/KQL/databricks, MCPs, semantic layer), including going fully self-hosted and OSS, and testing it all out on an AI CTF. Sign up early since seats are limited and we expect these to fill up fast. DM us to connect if you might be in Las Vegas that week as well. Beyond the 4-hour Las Vegas workshop, we’re also running our summer cohorts for the full 2-day version - 16 CPE credits - for both US and EU time zones. Dates and details in the first comment Dave Herrald Roberto Rodriguez Amin Karbasi David Bianco Dawn Song  #BSidesLV #AISOC #AIAgents #RAG #SecurityAutomation #AIforSecurity #OpenSourceSecurity #CyberTraining

Noema Atlas just launched: a peer-to-peer network for distributing local LLM weights, built on Iroh, free and open source under Apache-2.0. How it works: Every file is identified by its content hash and a signed manifest, so the same weights pulled from any peer dedupe into one verified copy, and every byte is checked as it streams in. Hugging Face and mirrors stay as an opt-in fallback, downloads fail over automatically when a source dies, and identical files are stored once via reflink/hardlink. You can also rescue and reshare weights that got pulled from HF. For those of us working in air-gapped and edge environments, content-addressed distribution with cryptographic verification is genuinely useful. When you're standing up inference-in-a-box inside a boundary, "where did these weights come from and are they exactly the bytes the publisher signed" is not a trivial question; it's the whole provenance story. A signed manifest plus per-byte hash checking gives you an answer you can actually audit, regardless of which peer or mirror served the file. Model availability shouldn't hinge on one host's uptime or takedown policy. If a model you depend on disappears from HF tomorrow, a P2P network where peers already hold verified copies means your supply doesn't evaporate with it. It's early and the team is explicitly asking for help, so it’s far from a finished product. But the design choices are the right ones: content addressing, signed manifests, machine-to-machine transfers over Iroh, native lightweight desktop app across macOS/Windows/Linux. #LocalLLM #OpenWeights #EdgeAI #Iroh #P2P #ModelProvenance #AirGap #AIEngineering

Black Hat Innovators & Investors Summit Emcee 🎤 If you've followed me for any time, you know that outside of being a security practitioner and enthusiast about the field of cybersecurity, I'm also very passionate about the startup, investment and innovation landscape. That's why I'm really looking forward to serving as the Emcee for Black Hat's Innovators & Investors Summit. This is the room where the next era of cybersecurity is being built and funded, from founders, investors and CISO's/security leaders. The agenda is shaping up awesome, and I'll have some big shoes to fill for prior emcee's such as my friend Mike Privette Be sure to check it out, as an opportunity to build relationships, engage with decision-makers, expand your network and understand the emerging security landscape and the teams defining it! https://lnkd.in/eEmHG_AY

More AI fun: I’ll be speaking at the SANS Cloud Security Exchange Summit on “Breaking Bots 2: Trusting Agentic AI in the Cloud SOC” When most orgs are putting LLMs and agents into things that matter like strategies for their AI SOC, their anti-fraud program, and overall trusted model, a lot of it boils down to trusting a number from an evaluation. The problem with testing AI is that the exams are now actively rigging themselves, and how to avoid getting fooled has gotten a lot trickier. We’re regularly seeing models answer from memorized results without touching a single log line, and even hacking out of their sandboxes to peek at the answers. In practice, agents under test are starting to act more like advanced persistent threats than predictable tools. I’ll walk through how to measure agents for what matters, what attacks against evals look like in the wild, and how to harden evals to actually hold up. If you’ll be in SF and are into agentic AI or graphs for investigations, reach out and let’s meet! If not, we’re running trainings on AI for investigations over at https://lnkd.in/gcpZu_CR , and happy to chat in general. SANS: https://lnkd.in/gNqwYHzv Joining a strong group of speakers this year: Anton Chuvakin Chris Cochran Eric Johnson Frank Kim Wesley (Wes) Kuzma Gee Rittenhouse Dave Shackleford Special thanks to: Dave Herrald Sarah Bird Joshua Saxe Ram Shankar Siva Kumar 🦝 #CloudExchangeSummit #AIsecurity #AgenticAI #AISafety #SOC

Finished franchise training last night in Jersey City with the great folks at Break Coffee Co. and now it's all official, and I'm excited to get home and get to servicing customers across Boulder and Denver metro!